(he/him)
Researcher in Applied Mathematics & Machine Learning
I recently graduated from the Master MVA with highest honours at École Normale Supérieure Paris-Saclay, and completed my research period at the École Normale Supérieure de Paris (IBENS, ENS Ulm).
My work focused on multi-marginal Schrödinger bridges, bridging probability theory, optimal transport, and deep generative modeling.
This research led to a paper submission to ICLR 2026, extending Schrödinger bridges to high-dimensional video and biological data.
I am now seeking a PhD position — academic or CIFRE — in machine learning theory, stochastic processes, or optimal transport, where I can further explore the intersection of rigorous mathematics and generative modeling.
Beyond research, I am passionate about sailing expeditions, paragliding, and mountaineering, constantly seeking exploration and challenge both intellectually and physically.
Master 2 MVA (Mathematics, Vision & Learning)
École Normale Supérieure Paris-Saclay
Master in Engineering
Arts et Métiers
CPGE PCSI-PC
Lycée Thiers, Marseille
My research lies at the intersection of optimal transport, diffusion models, and generative modeling. I am particularly interested in building both the theoretical foundations and practical algorithms for multi-marginal problems in machine learning.
Currently seeking a PhD position (CIFRE or academic) in applied mathematics and machine learning.
Here are several research and coding projects carried out during my year in the Master MVA (Mathematics, Vision, Learning) at ENS Paris-Saclay and my research period at ENS Ulm.
For a more exhaustive list, including technical reports and code, visit the GitHub repository:
github.com/tgravier/MVA-Portfolio
Novel approach for video generation using multi-marginal transport and Schrödinger bridge methods. Generates dynamic video sequences from static, unpaired image data through …
Segmentation of ultrasonic well imagery using transport and denoising architectures. Ranked 1st on both public and private leaderboards in the Collège de France EchoCem challenge.
Study and implementation of Monotone Gradient Networks (MGN) for learning gradients of convex functions, optimal transport, and generative modeling on high-dimensional data.
Reimplementation of Wave-U-Net for joint speech and noise separation directly in the time domain. Developed an end-to-end 1D U-Net architecture trained from scratch for robust …
Presentation and reproduction of PersLay (Carrière et al., 2020), a neural network layer designed to process persistence diagrams. Explored graph topology embeddings, heat kernel …
Reimplementation and theoretical study of Gradient Boosting as functional optimization, following Biau & Cadre (Annals of Statistics, 2021). Additional experiments explored …
Comparative study of generative and discriminative classifiers under adversarial and non-adversarial perturbations on medical imaging data. Implementation of GBZ and DBX models and …
Graph generation challenge using a Neural Graph Generator with diffusion and variational autoencoders. Our final model, based on SAGEConv and a modified decoder, achieved the 3rd …
